.png){kind=logo}
Faster generation: ChatGPT Images 2.5 cuts image-generation latency by up to 50% compared with Images 2.0, according to OpenAI.
Sketch and editing: The new Sketch feature turns rough drawings into finished images, while comments allow users to target specific areas for changes.
Two API models: GPT-Image-2.5 Flare focuses on fast everyday generation, while Sunburst targets detailed workflows where editing precision matters most.
OpenAI rolled out a new version of its image tool inside ChatGPT. It is called Images 2.5, and it brings faster image generation, sharper editing controls, and a few creative features that were missing before. The company calls it the most capable image model it built so far, one that can handle detailed instructions, work well with reference photos, and hold up across several rounds of edits.
Also Read: ChatGPT Images 2.5: OpenAI Boosts Editing Control With Sketch
Three things stand out in this update: better image quality, more accurate editing, and quicker generation. OpenAI says images now generate up to 50% faster than they did with the older Images 2.0 model. That alone changes how the tool feels to use. Lighting looks more natural, textures come out sharper, and reference photos hold their detail. A face uploaded as a reference, for instance, keeps its features intact even when placed in a completely different scene.
The changes are not limited to the model itself. OpenAI also added new tools inside the ChatGPT app. A Sketch feature lets people draw an idea before turning it into an image. New templates guide users through common formats like posters and product shots. A comment-style editing system lets people point at parts of an image and describe what to change. There is also a new way to share the exact prompt behind an image. All of this is free for existing subscribers, and it works across ChatGPT, ChatGPT Work, and Codex, on phones, computers, and the web.
Sketch is one of the features people are talking about the most. It gives users a small drawing canvas inside ChatGPT, where they can rough out an idea before asking the AI to build on it. Someone could draw the layout of a room, the outline of a jacket, or even a simple stick-figure doodle. After the sketch is done, a short written description is added to explain the style or the details that matter. ChatGPT then turns that rough drawing into a finished, polished image.
The point of the tool is simple: some ideas are easier to draw than to describe in words. A person does not need drawing skills to use it well, since the sketch only guides the shape and layout of the final picture. Once the image is ready, the usual editing tools are still available, so colours, backgrounds, or small details can be adjusted without starting over from the beginning.
Editing saw the biggest jump. A new toolbar appears whenever an image is opened, with five options: Markup, Comment, Remove Background, Erase, and Resize. Remove Background pulls the subject out and gives a transparent file, useful for slides or documents. Erase removes unwanted objects with a rough brush stroke, so exact precision is not needed. Resize switches between common formats like portrait, landscape, and widescreen.
The Comment tool is the one that feels genuinely new. A user can click directly on part of an image, type something like ‘remove this’ or ‘make it blue,’ and the AI acts on that one instruction alone. This moves editing away from typing long prompts again and again, toward small, direct clicks and notes. OpenAI says the model is now much better at changing only the part that was asked for, while leaving the rest of the image untouched.
One common complaint with older versions was quality dropping after several edits in the same conversation. A small change here, another there, and by the fifth or sixth edit, the image barely looked like the original. OpenAI says this version fixes that. Each new edit builds on the last one instead of slowly wearing the image down, so earlier changes stay in place even as new instructions get added.
This consistency also shows up in more complex work, like transparent backgrounds, infographics, and slide-style layouts. Teams working on a set of related images, such as a full ad campaign or a batch of social posts, should notice fewer surprises between one image and the next. That kind of steadiness matters more for marketing and content teams than for someone making a single picture.
Templates give people a starting point instead of a blank prompt box. Formats like posters, merchandise, and product photography now come with their own step-by-step guide. A person picks a template, answers a few quick questions about style and setting, and ChatGPT builds the image from those answers.
Sharing also changed with time. Along with the finished image, a user can now share the exact prompt used to make it. Someone else can take that same prompt, add their own photo or detail, and get a version that feels personal to them. It is a small addition, but it makes good ideas easier to pass around.
Developers get two new models through the API. Flare is the default choice, built for speed and lower cost, and it suits everyday tasks like social posts or product images at scale. Sunburst takes longer to generate an image, though it offers tighter control, which makes it a better fit for polished campaign work. Companies such as Adobe, Runway, and Manus have already started using these models in their own products.
Early reviews place Images 2.5 ahead of Google's Nano Banana 2 in a couple of areas, mainly editing accuracy and speed. Speed used to be a weak point for ChatGPT's image tools, so this is a real shift. What reviewers seem to like most is not the raw output quality, it is how the new editing toolbar and Sketch feature change the day-to-day process of making an image.
Nothing about this release feels like a complete rebuild. It reads more like a careful polish of a system that already worked reasonably well. For everyday users, faster generation and easier editing make the tool simpler to live with. For businesses, the new API options give a real choice between speed and precision, depending on the job at hand.
Also Read: ChatGPT Can Now Learn Your Personal Writing Style
Images 2.5 puts control ahead of raw generation. Sketch, the comment-based editing toolbar, and the new templates all point toward the same idea: give people a way to land on the exact result they had in mind, instead of hoping the AI gets it right on the first try. The speed increase matters too. Cutting wait time in half removes a lot of the friction that made earlier versions frustrating to use for repeated edits.
For casual users, this update turns ChatGPT into a more practical tool for small, everyday jobs, touching up a photo, putting together a poster, or making a quick product shot. For businesses and developers, the two API models offer a real choice between speed and precision, based on budget and the size of the job. On the whole, this feels like steady, sensible progress rather than a reinvention, and early reviews suggest it holds its own against rivals like Nano Banana 2, especially when it comes to editing and speed.
ChatGPT Images 2.5 is OpenAI’s latest image-generation model, offering sharper details, faster generation, improved reference-image fidelity and more reliable editing across multiple rounds of changes.
The Sketch tool lets users draw directly inside ChatGPT and use the drawing as a visual reference. Users can add a description, allowing the model to create a more complete image from the rough concept.
GPT-Image-2.5 Flare is designed for fast, high-quality everyday image generation, while Sunburst focuses on detailed creative workflows requiring greater editing precision and more control.
OpenAI says Images 2.5 is rolling out to ChatGPT, ChatGPT Work and Codex users across all tiers on desktop, mobile and web. ChatGPT Images is also listed as available across all ChatGPT tiers.
The updated model is designed to preserve important visual details across multiple edits. OpenAI says it follows editing instructions more reliably while retaining composition, reference details and the broader visual direction.
